Internal Links Need Jobs, Not Quotas: A 12-Case Editorial Routing Test

If a reader has to leave a useful page and search your site again to take the next sensible step, the link system has failed that reader. The repair is not “add five links to every post.” Give each link a job, name the reader’s next question, and connect only pages that can answer it. This is a practical editorial routing protocol for deciding whether a link belongs, where it belongs, what its anchor should promise, and how to verify that the destination is reachable.

For a concrete example of identifying pages with no editorial inbound links, see our SEO data we track. That guide addresses orphan discovery. The protocol here addresses a different decision: when two pages exist, is there a useful, honest reason for this page to point to that one?

The short rule is simple: add an internal link when it advances a reader’s task, supplies evidence, or clarifies a relationship that the surrounding text actually explains. Put it where the need arises. Use anchor text that describes the destination. Leave it out when the fit is weak, when it repeats a stronger route already present, or when the only reason is to satisfy a quota. Track discovery, usability, and business outcomes as separate questions. A link may help a person without changing a search result, and no single link change proves a ranking effect.

The decision this protocol resolves

A site editor commonly has three pages open: a source article, a possible destination, and a spreadsheet of URLs. The spreadsheet says that the destination has few incoming links. That fact can identify a candidate for review, but it cannot tell the editor what a reader needs. A page with few links may be intentionally narrow, obsolete, redundant, legally isolated, or already reachable through clear navigation. Conversely, a frequently linked page can still be difficult to use if links arrive with vague labels or in unrelated paragraphs.

The protocol answers one unit of work: a proposed source-page and destination-page pair. It does not score an entire website in the abstract. The editor writes down the source reader’s current question, the destination’s answer, the point in the source where the question arises, and what the anchor promises. If those four items cannot be stated plainly, the proposal is not ready to publish.

This framing prevents a common category error. “The destination needs more internal links” is a site-health observation. “A reader comparing crawl discovery methods needs the sitemap limitations explained here” is a link rationale. The first may generate a queue; the second can justify an actual editorial link. Queues are useful for prioritization, but they are not publishing instructions.

Internal links and outbound citations do different work

An internal link connects two URLs under the same site’s control. It can help readers move through a topic, find a relevant detail, or reach a next step; it can also make a destination discoverable through crawlable page links. Google’s link best-practices documentation describes links as a way Google discovers pages and assesses relevance, recommends ordinary crawlable anchors with an href, and advises descriptive, concise, contextual anchor text. It also says each page a site cares about should receive at least one link from another page on that site, while explicitly rejecting a magical ideal count for links on a page.

An outbound link points to another site. It may cite a source, let the reader verify a rule, or provide a tool or dataset the publisher does not own. Google’s documentation also says that linking out is not something to fear when it makes sense; its separate outbound-link qualification guidance explains when paid or user-generated links need qualification. Those are different decisions from whether two pages in your own content library belong together.

Do not make an internal link impersonate independent evidence. If the source is your company’s service page, say that it is your page and link because it is the relevant next step. If a claim comes from a public standard or a study, cite that source next to the claim. A reader should be able to tell whether a link is navigation, evidence, or a commercial offer without decoding your site architecture.

Why a quota is the wrong unit

A quota counts links; an editorial decision accounts for a relationship. If an editor is told to add a fixed number, the easiest path is to link headings, repeat the same destination, add generic “read more” anchors, or squeeze commercial pages into informational passages. The output may satisfy a spreadsheet and still make the page harder to scan. A quota can also reward link inflation while leaving the real routing problem—one useful page that nobody can find—untouched.

This does not mean link counts are useless. Counts can help locate anomalies, such as a destination that appears to have no in-content inbound links. They are a diagnostic signal, not a quality verdict. Count links after classifying them, and distinguish navigation, contextual editorial references, citations, product links, and repeated template links. A footer present on every page is not equivalent to a contextual explanation in a paragraph. A page can have a measurable link and still have no useful route for a reader’s question.

Google says there is no magical ideal number of links on a page. That is a sourced fact about the documented guidance, not a promise that any arbitrary link density is good. The editorial recommendation in this article is an inference from that guidance: choose a link because the reader’s task calls for it, then inspect the page for clutter and repetition. The right count is an outcome of the content, not a target imposed before reading it.

What the available evidence does and does not show

The evidence base is thinner than many “orphan page” statistics make it appear. One publisher’s audit of 153 posts reports that 29 percent of the pages in its own blog sample met its orphan-page definition. A separate SaaS blog audit describes 15 sites, 10,426 posts, and 44,513 in-content links, reporting 38.2 percent of sampled posts with zero in-content inbound links. The authors use their own site selections and operational definitions. These are useful examples of how audits can be designed; they are not a representative estimate of all websites, and their percentages should not be compared as though the samples and definitions were identical.

The first-party and multi-site audits are observational. They show what those auditors found in selected content sets, not that adding a link caused a page to be crawled, indexed, visited, or ranked. Differences in site selection, URL eligibility, what counts as an in-content link, pagination, canonicalization, and crawl scope can materially change the count. Treat the percentages as prompts to inspect your own data, never as a universal failure threshold.

A practitioner question on how editors choose which related pages to link illustrates the operational uncertainty: people ask whether to link a related post, a service page, or both, and commenters offer different rules of thumb. That discussion is anecdotal demand evidence, not a controlled study. It is useful because the uncertainty is about the editorial decision itself, which raw site-wide counts do not settle.

Consulting guidance adds a broader content-operations lens, with a limit. McKinsey’s digital-marketing article describes content as part of a customer decision journey and warns that poorly linked material can be hard to find in a large publishing operation. The article is from 2010 and concerns enterprise marketing operations rather than a current search-ranking experiment. Use it to support the organizational point—content routes should make useful material findable—not to claim a measured SEO effect or a modern ranking factor.

The route test: four questions before insertion

For each proposed link, write four short answers. First, what is the reader trying to understand or do in the paragraph containing the link? Second, what specific answer, evidence, or action will the destination provide? Third, why is this the right moment to offer that destination rather than a different one? Fourth, does the visible anchor set an accurate expectation about the destination?

A good answer is concrete: “The paragraph explains why an XML sitemap does not guarantee indexing; the destination shows how to inspect crawlable internal paths; the anchor names that inspection guide; the link belongs here because the reader is choosing a discovery check.” A weak answer is “related topic,” “SEO authority,” or “we need another link.” The weak answer may signal a page relationship worth investigating, but it does not yet establish a reader benefit.

The four answers also expose commercial intent. If the destination is a product or service, the source passage should make the commercial transition understandable. A tutorial on diagnosing missed calls can link to a service offer after explaining what that service does and who it suits. It should not interrupt a neutral definition simply because the offer needs traffic. A transparent offer can be useful; an unexplained offer is a surprise.

A 12-case acceptance matrix

The following matrix is a designed editorial test suite, not a measured study of user behaviour. Its sample size is 12 hypothetical source–destination pairs selected to cover common publishing decisions. The suite was drafted on 26 October 2026. It excludes analytics experiments, crawler log analysis, accessibility testing with users, and any claim that passing cases improve ranking. Its purpose is reproducibility: two editors can review the same candidate pair and explain why they accepted, revised, or rejected it.

Case	Source situation	Proposed destination	Decision	What must be true
1	A paragraph defines a technical term	A page with the full explanation	Accept	The anchor names the term or guide; the destination expands it without changing the claim
2	A process lists a decision step	The exact next-step checklist	Accept	The checklist begins where the source leaves off and is current
3	A claim relies on a public rule	The primary authority’s rule page	Cite	The link supports the claim and does not substitute a company page for the authority
4	A post mentions a tool category	A vendor landing page	Reject or disclose	A commercial relationship or recommendation is clear; the source passage has a genuine buying purpose
5	An introductory paragraph mentions “SEO”	A generic SEO services page	Reject	Broad topic overlap alone does not explain why this destination helps this reader now
6	A page gives an example with a limitation	A deeper page that explains that limitation	Accept	The anchor signals the limitation; the destination actually resolves it
7	A long guide lists ten subtopics	Ten links in one dense list	Revise	Group by reader task, use descriptive labels, and retain only destinations that add distinct value
8	A page repeats the same destination in every section	The same destination repeated	Usually reject	Keep the clearest useful occurrence unless separate contexts genuinely require different routes
9	An old post references a renamed page	The current canonical destination	Accept after validation	Confirm redirect, canonical, status, and that the content still matches the reference
10	A destination has no in-content inbound link	A source page with a genuine relevant passage	Candidate	Add only after the four-question route test; zero count alone is insufficient
11	A source page is about one country	A destination about another jurisdiction	Reject or qualify	Make the jurisdiction change explicit and ensure the destination is not misleading
12	A source page is informational	A commercial next step	Conditional	Explain the transition and audience; don’t disguise promotion as evidence

Cases 1–3 separate explanation, workflow, and evidence roles. Case 3 is especially important: a citation to a primary source is not simply another internal-navigation edge. Cases 4 and 12 deal with commercial destinations, where the reader may still benefit from a next step but needs a candid signal of purpose. Cases 5 and 10 block weak reasons: adjacency and a zero count are not proof of fit.

Cases 7 and 8 address density and repetition. More destinations can be appropriate in a directory or comparison table, but a prose paragraph with many adjacent links can become difficult to read. Google’s guidance specifically cautions against chaining links together and recommends context around them. The solution is not a numeric cap; it is to regroup a list, explain the choices, or remove links that do not serve a distinct job.

Cases 9 and 11 protect accuracy over time and across jurisdictions. A previously valid link can become stale when a URL changes, a page is retired, or a destination changes scope. An otherwise relevant link can mislead if an article about one country points to a rule for another without saying so. Link audits therefore need content review, not only HTTP status checks.

Score a candidate without pretending the score is truth

A lightweight rubric can make reviews consistent. Rate each dimension from zero to two: reader-task fit, destination specificity, placement, anchor accuracy, freshness, and non-duplication. Zero means absent or misleading; one means partly supported; two means clear and verified. A total of 10–12 can be accepted, 7–9 should be revised, and 0–6 should normally be rejected. These cutoffs are a proposed editorial control, not a validated predictive model. They should be adjusted after editors compare disagreements and observe usability feedback.

Use hard stops in addition to the sum. Reject a link if it resolves to the wrong topic, a broken page, a page that is not intended for public access, or a destination whose claims conflict with the sentence. Pause if the URL redirects to a materially different topic or if the page’s canonical points elsewhere. A high average score must not mask a serious failure in truthfulness or destination identity.

A practical review record can be short: source URL, destination URL, role, paragraph location, anchor, the reader question, status code, canonical destination, reviewer, decision, and reason. Keep the text of the reason. A spreadsheet with a “score 11” and no explanation cannot tell a future editor whether the link was a citation, a tutorial route, or a conversion step.

Placement: link where the question appears

Place the link where a reader encounters the need it satisfies. If a paragraph introduces a method and the next paragraph explains the method in depth, the link belongs near that transition. If the destination is a primary source, put it beside the factual claim, not in a generic “resources” list after several unrelated sections. If it is a commercial next step, place it after the reader has enough context to decide whether the offer fits.

Do not force every link into the opening. An early link is useful when it answers an immediate question, but a link near the relevant technical term may be clearer. The goal is visible, contextual navigation, not front-loading. On long pages, a table of contents, section links, and breadcrumbs may make the broader structure easier to navigate; these structural mechanisms complement contextual links instead of replacing every one.

Placement should also work for scanning. A link embedded in an overly long sentence can be missed. A link with a vague anchor such as “here” makes the reader inspect surrounding text to infer the destination. Google recommends concise, descriptive text that remains meaningful when read out of context. A useful editor test is to read the anchors alone: can a reader predict the destination without guessing?

Anchor text: promise only what the destination delivers

An anchor is a small promise. “How to fix orphan pages” promises a practical repair guide. “Search documentation” promises official documentation. “Our SEO services” signals a commercial destination. Do not use a promising, educational anchor when the destination is a sales page, and do not use a keyword-heavy phrase that no normal editor would write in that sentence.

Vary anchors naturally when several routes point to one destination, but keep the expectation accurate. Exact repetition is not a requirement. Nor should editors stuff every variant of a keyword into different paragraphs. The reader needs to know what the link is, not decode a ranking tactic. Google’s own examples advise against generic labels and unnaturally long link text; they also say surrounding context matters.

Check the destination after drafting the anchor. Read its title, opening, scope, and date. If the page no longer fulfills the anchor’s promise, update the anchor or choose another destination. An internal route can be factually wrong even when it returns HTTP 200. A crawler can verify reachability; it cannot determine whether “implementation checklist” accurately describes a page that has become a product comparison.

Discovery is not indexing, and indexing is not ranking

A crawlable internal link can help a search engine discover a URL, but that does not guarantee a crawl at a particular time, inclusion in the index, or a ranking. Google’s sitemap documentation likewise says submitting a sitemap is a hint and does not guarantee that URLs will be crawled or indexed. Sitemap presence and editorial links answer related but different questions: a sitemap lists URLs a publisher wants considered; a contextual link gives a reader and crawler a route from one page to another.

If a page is absent from search, do not diagnose “needs more internal links” from a link count alone. Check whether it is indexable, canonicalized as intended, accessible to crawlers, useful and distinct, and discoverable through navigation or links. Inspect server responses, robots directives, rendered anchors, canonical signals, and Search Console indexing information. Then decide if a contextual link is warranted. If the source text has no natural reason to point there, improve the information architecture or the destination’s role rather than inserting a sham link.

Conversely, a page can be indexed and still be difficult for readers to reach. Search visibility does not replace usable navigation. Internal linking is part of information architecture and reader flow, not a switch that guarantees an algorithmic result. Report the evidence at the level it supports: “we added a crawlable route,” “the destination gained a link from this page,” or “users selected the route in our measured session sample.” Do not rewrite these as “this link caused ranking growth” without a valid causal design.

A small executable audit for candidate links

The following Python script checks a CSV of proposed routes for duplicate pairs, empty anchors, anchors that are too generic, and the presence of a written rationale. It does not fetch pages, render JavaScript, test accessibility, confirm canonical tags, or judge whether a relationship is useful. Those checks require separate review. The point is to make administrative defects visible before publication, not to automate editorial judgment.

CSV columns: source_url,destination_url,anchor,role,reader_question,rationale. Save as routes.csv, then run python audit_routes.py routes.csv.

import csv
import sys
from collections import Counter

GENERIC = {"here", "click here", "read more", "this page", "learn more"}
REQUIRED = {"source_url", "destination_url", "anchor", "role", "reader_question", "rationale"}


def main(path):
    with open(path, newline="", encoding="utf-8-sig") as f:
        rows = list(csv.DictReader(f))
    if not rows:
        raise SystemExit("No route rows found.")
    missing = REQUIRED - set(rows[0])
    if missing:
        raise SystemExit(f"Missing columns: {', '.join(sorted(missing))}")

    pairs = Counter((r["source_url"].strip(), r["destination_url"].strip()) for r in rows)
    errors = []
    for line, row in enumerate(rows, start=2):
        anchor = " ".join(row["anchor"].split()).strip().lower()
        if not row["source_url"].strip() or not row["destination_url"].strip():
            errors.append(f"line {line}: source or destination is blank")
        if anchor in GENERIC:
            errors.append(f"line {line}: generic anchor {anchor!r}")
        if not row["reader_question"].strip():
            errors.append(f"line {line}: reader question is missing")
        if not row["rationale"].strip():
            errors.append(f"line {line}: rationale is missing")
        pair = (row["source_url"].strip(), row["destination_url"].strip())
        if pairs[pair] > 1:
            errors.append(f"line {line}: duplicate source/destination pair")
        if row["role"].strip().lower() not in {"navigation", "explanation", "evidence", "commercial", "tool"}:
            errors.append(f"line {line}: unknown link role")

    if errors:
        print("REVIEW REQUIRED")
        print("\n".join(f"- {e}" for e in errors))
        raise SystemExit(1)
    print(f"STRUCTURE OK: {len(rows)} candidate routes. Editorial review still required.")


if __name__ == "__main__":
    if len(sys.argv) != 2:
        raise SystemExit("Usage: python audit_routes.py routes.csv")
    main(sys.argv[1])

The script intentionally does not assign a “quality score.” A CSV cannot tell whether a phrase is relevant, whether the destination is current, or whether the link interrupts a sensitive explanation. It can prevent a blank rationale and repeated pair from slipping into a content ticket. A human still needs to open both pages and check the context.

Measure the change at the right level

If the objective is findability, record the destination’s eligible source pages, discovered internal links, crawl status, and observed search-engine crawl dates where logs are available. If the objective is reader navigation, define a specific event such as a click on the contextual route, then compare the rate with an appropriate baseline. If the objective is a commercial outcome, measure the route’s qualified next-step completion, not raw click volume alone. These are separate outcomes and should not be collapsed into one “SEO score.”

A simple before-and-after count is descriptive. It cannot rule out seasonality, content changes, promotions, changes in crawl demand, or a concurrent site migration. A matched comparison can reduce some confounding if comparable pages are selected in advance, but it still depends on matching quality and stable measurement. A randomized test may be possible for user-facing placements on sufficiently high-traffic pages, but avoid changing content that should remain consistent for accessibility, legal accuracy, or the user’s task. Document what changed and what did not.

A practical measurement record includes the test date, source page and destination, old and new anchor, placement, publication timestamp, release or crawl status, eligible sessions or impressions, outcome definition, comparison group, exclusions, and known concurrent changes. Avoid interpreting low counts as evidence of “no effect.” If the sample is small or data is noisy, report the uncertainty and keep the decision reversible.

For search-console data, set a fixed observation window and compare like with like. Search Console performance data is affected by query mix, country, device, search appearance, and reporting windows. Do not attribute a change to a single link when titles, internal navigation, content, templates, or external demand changed too. If you have no credible counterfactual, report the observation as an operational signal, not a causal result.

Failure modes and their repair

The link exists only in a menu or footer. Template links can be valid navigation, but they may not explain the contextual relationship in the body. Repair the information architecture if that is the intended route; add an editorial cross-reference only when the paragraph supports it.

The anchor is “read more.” This hides the destination’s value from scanners and assistive technology users. Replace it with a concise description of the target, or remove the link if the destination cannot be described naturally.

The destination is a sales page disguised as evidence. Separate source evidence from offers. Label the commercial next step plainly and disclose the relationship where policy requires it. The Federal Trade Commission’s business guidance explains that advertising and endorsements must not mislead consumers; apply the specific rules relevant to your jurisdiction and relationship.

A link points to a page that redirects, errors, or changes topic. Check the final URL, response, canonical, and page content. Fix the target or remove the route. A successful HTTP status does not certify topic fit.

The editorial team adds links only to destinations with poor metrics. This turns an operational list into a quota. Keep the candidate queue, but require the route test and record “no relevant source paragraph” as a valid result.

The team celebrates a ranking change as proof. Search results vary, and multiple site changes can coincide. State what was observed, what was controlled, and what remains unknown. Do not promise that a new link will rank a page.

The internal link is repeated as an external citation. Internal navigation and independent evidence have different trust roles. Cite the primary authority or study near the claim; use your own page for additional explanation or a clearly marked offer.

A review checklist editors can use

Before accepting a proposed link, confirm all of the following:

• The reader’s question in the source paragraph is explicit.
• The destination answers that question or gives a transparent next action.
• The anchor describes the destination accurately and concisely.
• The link is placed where the need arises, with surrounding context.
• The destination is public, reachable, current, and canonicalized as intended.
• The destination is not a near-duplicate, obsolete page, or wrong jurisdiction.
• The link has one clear role: navigation, explanation, evidence, tool, or commercial next step.
• A claim’s supporting source is cited independently when needed.
• Paid, sponsored, or user-generated links are qualified when applicable.
• The rationale is recorded so another editor can understand the decision.
• Any measurement plan separates discovery, reader behavior, and business outcomes.
• The team has not treated the proposed score or link count as a ranking guarantee.

For a team, use the same form in the CMS ticket: source URL; destination URL; paragraph; reader question; anchor; role; evidence; freshness check; decision; reviewer; and date. This makes the route auditable and makes “no link” a legitimate editorial decision rather than a failed quota.

A Mermaid view of the review flow

flowchart TD
  A[Candidate source and destination] --> B{Can we state the reader question?}
  B -- No --> X[Reject or reframe the content task]
  B -- Yes --> C{Does the destination answer it?}
  C -- No --> X
  C -- Yes --> D{Is the role transparent?}
  D -- No --> Y[Rewrite anchor or disclose commercial role]
  D -- Yes --> E[Check status, canonical, freshness, scope]
  E --> F{Destination still fits?}
  F -- No --> Z[Repair target or remove link]
  F -- Yes --> G[Place link in context and record rationale]
  G --> H[Measure the intended outcome separately]

This flow deliberately contains rejection branches. An editorial process that can only approve links will eventually manufacture weak ones. The best outcome for some candidate pairs is to improve the source content, create a genuinely needed destination, or leave the pages unlinked because their tasks differ.

What changes when the site is large

Large sites need automation to produce candidate pairs, but candidate generation is not approval. A crawler can find pages with zero incoming links, broken destinations, redirect chains, duplicate anchors, deep click paths, or links that appear only in templates. Search Console can provide indexing and performance clues. A content inventory can add topic, owner, audience, publication date, freshness, and business role. These signals help people focus the queue.

The review still needs a human-readable reason. A page with zero inbound links may have a valid route in a product interface that a public crawler cannot access. A high-value destination may intentionally require authentication. A near-duplicate page may need consolidation, not additional links. A legacy page may be kept for compliance or historical reasons but should not be promoted as current guidance. Before changing links at scale, define which URL is canonical, which is maintained, and which is retired.

For repeated templates, separate site-wide navigation from editorial links in the data. Otherwise, an audit may report thousands of inbound links that all come from the same footer and hide the fact that no article explains the destination. Conversely, a templated link may be exactly the correct mechanism for a category page. The metrics need to reflect the information architecture you are actually trying to improve.

A simple queue should include: candidate destination; eligible source pages; source paragraph; proposed reader task; link role; indexability and canonical status; owner; review state; and result. Sort by reader value and implementation effort, not by the largest raw count. If two destinations answer the same question, choose the maintained, authoritative one or consolidate the material before adding more routes.

When a link is not the fix

Sometimes the reader needs a better destination, not another path. If the destination is thin, outdated, or ambiguous, sending more people there can amplify confusion. Update it first or merge it into a stronger page. If two pages target the same task, clarify their different roles or consolidate them. If the only relevant destination is behind a form, decide whether the route should set expectations before the reader clicks.

Sometimes the source page itself has no suitable paragraph. Do not insert a sentence solely to host a link. That is a sign that the source and destination might not have a meaningful relationship, or that the source needs a substantive section explaining the shared topic. Editors should not create filler to justify a link. A “related content” module can still work when its label is explicit and the recommendation is actually related, but it should not be used to disguise an unrelated promotion.

Sometimes discovery is blocked by technical conditions. A URL that is noindex, blocked by robots, canonicalized elsewhere, inaccessible without a session, or returning inconsistent responses needs technical diagnosis. Internal links cannot override those signals. The route may still help a human if the page is intended for use, but do not report it as an indexing fix until the technical cause is understood.

Sometimes the right route is navigation, not an in-paragraph link. A topic hub, breadcrumb, category structure, or product menu can express a stable relationship better than adding the same link to dozens of articles. Use editorial links for local context and structural navigation for durable hierarchy. The two can coexist, but each should have an explicit purpose.

A controlled rollout that avoids false certainty

Start with a small set of eligible page pairs that have clear reader tasks. Record the page versions, link role, anchor, placement, date, and outcome you expect. Ask an editor who did not create the link proposal to review a sample of decisions; disagreement is useful evidence that the rubric or source pages need clarification. If reviewers disagree about the destination’s promise, fix the anchor or the destination before publishing.

Separate implementation verification from outcome measurement. Implementation verification asks whether the intended link rendered as a real anchor, resolves to the intended canonical URL, and appears in the expected location across devices. Outcome measurement asks whether readers clicked, whether the destination received eligible visits, whether a search crawler discovered it, or whether a business outcome changed. These are different tests. A successful implementation does not imply a successful outcome.

If you compare treated and untreated pages, select the comparison before seeing the result. Match on topic, age, prior traffic, template, and page role where possible. Log title edits, external campaigns, site releases, and content updates during the observation period. The comparison will still have limitations, but recording them is better than retrofitting an explanation after a chart moves.

Do not force a long observation window to prove an effect. The appropriate period depends on crawl cadence, traffic volume, seasonality, and the decision’s cost. For a broken link, repair immediately and verify. For editorial placement choices, collect enough observations to make a decision without claiming certainty the data cannot support. If the evidence remains inconclusive, the team can still choose the more useful route for readers and label the search impact unknown.

Sources, facts, inferences, and recommendations

Sourced facts: Google documents crawlable link markup, descriptive contextual anchors, internal cross-references, and the absence of a magical ideal link count. It also says sitemap submission does not guarantee crawl or indexing. The two audits cited above report findings for their own selected samples and definitions. McKinsey’s article describes organizational content findability in an enterprise marketing context.

Inference: A zero inbound-link count is useful for finding candidates but is not sufficient evidence that a page should be linked. This follows because a count does not encode the reader’s task, destination fit, scope, maintenance state, or other technical signals. The independent audits do not establish a universal threshold or causal impact.

Recommendation: Use the four-question route test, the twelve designed cases, and a short review record. Pilot the rubric, inspect editor disagreements, and measure the specific outcome that matters. Do not claim that a route causes rankings to rise unless a credible experiment supports that causal statement.

Limitations of this protocol

The 12-case matrix is an editorial test suite, not an empirical sample. Its sample size is 12 designed scenarios; the date is 26 October 2026; there are no participant observations, click logs, crawl logs, or ranking measurements in the matrix. It excludes tests of screen-reader usability and automated rendering. The rubric and score cutoffs are proposed operating controls. A publisher should revise them after observing real editorial disagreements and user needs.

The cited orphan-page audits are self-reported publisher studies. They use different selected site sets and definitions; the reported percentages are not comparable prevalence estimates and cannot establish why the pages lacked links. McKinsey’s cited article is from 2010 and supports only a broad publishing-operations point. Google’s guidance is authoritative for Google’s documented advice, but its statements are not a guarantee of outcomes in a particular site or search result.

Most importantly, a useful route may not be measurable with sufficient power on a low-traffic site. “No measured lift” may mean no effect, insufficient data, noisy data, or a poor metric. Keep the distinction visible. The editorial question can be answered on user-value grounds even when the ranking effect is unknown.

The practical rule to keep

Every proposed internal link should be explainable in one sentence from the reader’s perspective: “At this point, a reader may need X, and this destination gives them Y.” If that sentence is accurate, the destination is current, the anchor describes it, and the link appears where it helps, publish it and record the reason. If the sentence cannot be written, do not fill the gap with a quota.

After publication, verify the route in the rendered page, then record only what you actually observed. A crawlable link can make another page easier to discover; it cannot guarantee crawl timing, indexing, ranking, user trust, or conversion. The durable benefit is a more legible site: readers can move to the next answer without guessing why the publisher sent them there.

Editor’s final checklist

• Is there one explicit reader question for this source passage?
• Does the destination answer it with current, useful material?
• Does the anchor describe the target rather than merely signal “click”?
• Is the link close to the need, but not awkwardly forced into the opening?
• Is this contextual reference distinct from a citation, navigation item, or offer?
• Does the destination return successfully and retain the expected canonical and scope?
• Are external factual claims supported by independent sources near the claim?
• Is the planned measurement about implementation, reader behaviour, discovery, or business outcome—and only one at a time?
• Have concurrent changes and limitations been recorded?
• Can the editor say “no link” when the relationship is weak?

When all answers are clear, the team has a defensible editorial decision. When they are not, the next step is to improve the source, destination, or information architecture—not to increase a link quota.